Simulate before you execute
Rehearse pressure before release. Stress-test demand, fraud, edge cases, and escalation while the consequences are still synthetic.
A book by Jesper Lowgren
Designing systems that can act.
Agentic System Design is a field guide for the moment software stops merely executing instructions and begins sensing, deciding, learning, and cooperating under constraints.
Policy
Trust
Proof
We have crossed a line. Software no longer waits.
Once intelligence can act, the design laws change.
This book is about those laws: how to shape agentic systems that perceive, decide, cooperate, and remain accountable as capability grows.
Most AI projects fail because they automate tasks without designing the conditions for autonomous behavior. They process data but do not bind meaning, policy, trust, and proof to action.
Agentic System Design reframes the work: from shipping features to shaping behavior, from isolated pilots to systems that improve under constraint, from retrospective governance to controls that run at the speed of the agents.
The core inversion
The old playbook broke.
Traditional design starts with outcomes, specifies processes, and audits later. Agentic design begins inside the system: with boundaries, signals, policies, proof, and the conditions under which adaptation remains legitimate.
Signals What the system can sense
Policy What must hold at runtime
Proof What can be explained later
Boundaries Where autonomy stops
Five shifts
From control to governed adaptation.
Design conditions, not outcomes
Autonomous behavior cannot be fully scripted. Shape the rules, signals, incentives, and guardrails where acceptable outcomes can emerge.
Architect for emergence
Surprise is not a bug in agentic systems. Build modularity, provenance, redundancy, and thresholds that absorb it without collapse.
Design for loss of control
Granular oversight will not scale. Prepare kill switches, graceful degradation, reversible escalation, and policies that run.
Govern what you cannot predict
Governance must happen as action unfolds: continuous provenance, live trust metrics, and distributed human intervention.
The system view
Agents need more than intelligence.
They need roles, passports, shared meaning, runtime controls, simulation evidence, human participation, and a data backbone strong enough to make every action traceable.
Agentic Data Backbone
Verifier
Planner
Executor
Coordinator
Policy
Human
Agent archetypes
Nine roles give agents atomic scope: perceptual, verifier, analytical, planner, decision, executor, coordinator, advisory, and learner.
Agent passports
Every event carries identity, authority, provenance, confidence, policy bindings, and lifecycle evidence.
Activation gates
Agents wake only when signals, thresholds, readiness criteria, and autonomy boundaries align.
Governance in motion
Policy-as-data moves governance from documents into runtime constraints, thresholds, and escalation logic.
Twelve chapters
A roadmap from mindset to governed autonomy.
The Inversion of Design
Why autonomy turns design inside out.
From Automation to Agency
The maturity logic behind agentic readiness.
From Inversion to Authopy
How cognition becomes executable and accountable.
Agent Archetypes
Role design for coherent multi-agent work.
The Data Backbone
Events, passports, provenance, trust, and shared meaning.
Activation Patterns
Signals, triggers, autonomy gates, and choreography.
Governance-by-Design
Policy-as-data and runtime control.
Simulation Design
Scenarios, sandboxes, failure, and rehearsal.
Human Participation
Beyond human-in-the-loop toward designed stewardship.
Readiness & Scaling
Resilience, culture, and failure modes at scale.
E-commerce Case Study
The full system in a concrete operating context.
The Road Ahead
Adaptive governance, learning loops, and the social contract.
Agentic readiness
Capability is only one quadrant.
The book treats readiness as a balance across capability, governance, trust, and human participation. Push one dimension too far ahead of the others and the system becomes brittle.
64Balance score
Who it is for
For the people who will shape the systems that shape us.
Enterprise architects
Design the reference models, operating patterns, and boundaries for agentic ecosystems.
Product and service designers
Move from screens and journeys to conditions, signals, and participatory control surfaces.
AI builders
Turn prototypes into traceable systems with role clarity, evaluation, provenance, and runtime policy.
Leaders and stewards
Understand where control changes, where legitimacy comes from, and what readiness really requires.
The proposition
Autonomy should expand control, not erode it.
Agentic System Design is the starting point of the Agentic Trilogy: a practical, rigorous way to build intelligent systems that can explain themselves, adapt responsibly, and operate at machine speed without losing human judgment.